HAT
=============

HAT is a parameter-isolation-based continual learning method which learns real-valued masks, which play a role similar to attention modules, for weighting each layer output for each task.
For the details, see the `original paper <https://arxiv.org/pdf/1801.01423>`_.


Node-level Problems
-------------------------

.. automodule:: begin.algorithms.hat.nodes
    :undoc-members:
    :members:
    :private-members: